Automatic Detection of Abnormalities Associated with ...€¦ · thresholding fails when background illumination is uneven. Multiple thresholds are used to avoid uneven illumination

International Journal of Computer Applications (0975 – 8887)

Volume 140 – No.4, April 2016

1

Automatic Detection of Abnormalities Associated with Abdomen and Liver Images: A Survey on Segmentation

Methods

G.G. Rajput Department of Computer Science

Rani Channamma University Belagavi, India

Anand M. Chavan Department of Computer Science

Solapur University Solapur, India

ABSTRACT Image segmentation plays an important role in medical

imaging by automating detection of false structures and other

regions of interest. An image segmentation method partitions

an image into multiple segments, representing an image into

more meaningful, simpler and easier to analyze. Several

general-purpose algorithm and techniques have been

developed for image segmentation. This paper explains

different segmentation techniques used in medical image

analysis addressing the segmentation of abdominal and liver

images as case study. Experiments are performed on

abdominal and liver CT scan images and the outcomes of

these segmentation techniques are discussed. Performance of

the methods is presented on the basis of parameters namely,

pixel values, mean and standard deviation.

Keywords

Segmentation, thresholding, clustering, artificial neural

network, edge detection, region of interest

1. INTRODUCTION Digital image processing is the use of computer algorithms for

processing digital images. One important application of digital

image processing is medical imaging. It is an important tool in

medicine. Computer Tomography (CT), Magnetic Resonance

Imaging (MRI), Ultra Sound (US) and other imaging technique

provides information about human anatomy. These

technologies are more useful in diagnosis and treatment

planning in medicine. Various methods can be applied to

images obtained from these technologies to obtain essential

features of images that help in diagnosis and treatment

planning. Image segmentation is one such popular method in

many image processing tasks. Segmentation is the process that

segregates the pixels to bring out objects of interest in an

image. A great variety of segmentations methods has been

proposed in past decade [2, 3, 30]. Image segmentation plays

an important role in biomedical-imaging applications. Medical

applications of the segmentation is the study of human

anatomy, to detect the region of interest, tumor burden etc.

There are different problems in the segmentation of medical

images such as intensity inhomogeneity, partial volume effect,

artifacts, and closeness in grey level of different tissue [27].

Generally, medical images are complex in nature (diverse

image content, cluttered objects, occlusion on-uniform object

texture) and noisy, hence, making it difficult for segmentation.

Further, boundary insufficiencies (i.e. missing edges and/or

lack of texture contrast between regions of interest (ROIs) and

background) make the segmentation task challenging. In this

paper we present a brief overview of current segmentation

methods [31] specifically used in detecting abnormalities in

liver and abdomen CT scan images.

Several authors [6] have brought out survey of various

segmentation techniques. However, comparative study of these

methods has not been uniformly presented using a single

database of medical images [7] thereby making it difficult for

evaluation and comparison. The aim of this paper is to review

segmentation techniques appeared in the recent literature on

medical image segmentation and compare their performance by

experimenting on liver and abdomen CT scan images of local

database collected from PRISM Diagnostic Center, Solapur.

Section II describes various image segmentation techniques in

brief. Section III focuses on the experimental results of various

segmentation algorithms. Finally, section IV presents

conclusion.

2. IMAGE SEGMENTATION AND

CURRENT SEGMENTATION

TECHNIQUES Image segmentation is the process of partitioning a digital

image into multiple segments [2]. When applied to medical

images, segmentation techniques identify the boundaries of

objects such as organs or abnormal regions (e.g. tumors) in

images. The resulting images enable for shape analysis,

detecting volume change, and making a precise radiation

therapy treatment plan. Various theoretical frameworks have

been proposed for segmentation. Among some of the popular

methods are Thresholding (Histogram thresholding and slicing

techniques) [11,12], region growing (starting in the middle of

an object and then “growing” outward until it meets the object

boundaries) [10,14], edge detection and grouping (detected

edges in an image are assumed to represent object boundaries)

[ 6,7 ], active contour models and level sets (PDE frame work)

[16,17], graph cut [ 32 ], and clustering [22, 23, 25, 26] (group

together patterns that are similar in some sense.) Significant

extensions and integrations of these frameworks [27] improve

their efficiency, applicability and accuracy. The choice of

segmentation techniques totally depends on the basis of type of

image and type of problem domain being considered so that the

results are acceptable by medical experts.

Different segmentation techniques are available in literature

but not a single technique is useful for different types of

images [4]. It is very difficult to develop a unified approach for

image segmentation of different types of images and also it is

very difficult to choose specific technique for a specific type of

image. Therefore, image segmentation is a challenging

problem in image processing. Due to different techniques,

image segmentation is broadly classified in to two categories,

on the basis of two properties of intensity values.

Detecting Discontinuities: This method is based on the

intensity variation among the pixels [1]. Any significant

changes in the intensity levels among neighboring pixels are

termed as edges and results in the discontinuity in the pixels. It

includes edge detection algorithms.

Detecting Similarities:-To partition an image in to regions on

the basis of predefined criteria [1]. This method works on the

https://en.wikipedia.org/wiki/Digital_image

https://en.wikipedia.org/wiki/Digital_image



2

basis of homogeneity criteria i.e. all pixels inside a region

possess similar characteristics and dissimilar to the pixels in

other region. It includes thresholding, region growing, region

splitting and merging algorithms.

2.1 Thresholding Thresholding is a simplest approach of segmentation having

light object on dark background [1]. This method is based on

the pixel values and image space region i.e. characteristics of

images [7]. Threshold based technique divides the image into

two classes; pixel belonging to certain range of intensity values

represents one class and the rest of the pixels in the image

represents the other class. It converts a multilevel intensity

image into a binary image i.e. it choose a threshold T, to divide

image pixel into several regions and separate objects from

background. If any pixel having intensity value is greater than

or equal to threshold value then these pixels are considered as a

part of object. i.e. f(x, y) ≥ T, else pixel belong to background

[10,11] . Mathematically, thresholding is represented as below

f

=

Where is the pixel value at the position .

Two variations of thresholding are common, global

thresholding and local thresholding [12]. When the value of T

is constant for the entire span of the image, thresholding is

termed global, otherwise it is local thresholding. Global

thresholding fails when background illumination is uneven.

Multiple thresholds are used to avoid uneven illumination in

local thresholding [9]. Thresholding does not yield results for

multichannel images, and is sensitive to noise and intensity

inhomogenities [7].

2.2 Segmentation based on Edge Detection Edge detection methods are based on intensity variations

among the pixels. Intensity data of an image provides

uncertainly information about the location of edges [5]. This

technique will resolve image segmentation by detecting edges

between different regions that have discontinuity in intensity is

extracted [1, 6]. The boundaries of the two or more objects

forms edges and major changes in the intensity levels between

neighboring pixels in a certain direction are termed as edges

and results in the discontinuity in the pixels. The resultant

image is a binary image [7, 8]. Uncertainty of edges occurs

because of noise introduced in the imaging process and later in

the transmission and sampling process. The simplest way to

detect edges is to look for abrupt intensity change using first

derivative and second derivative of the intensity. There are two

main edge based segmentation methods [7].

2.2.1 Gray Histogram Techniques The final result of edge detection method mainly depends on

the value of threshold T. It is very difficult to detect maximum

and minimum gray histogram due to uneven impact of noise.

Therefore, this method substitutes the curves of object and

background with two conic Gaussian curves [7].

2.2.2 Gradient based method The edge based techniques are mainly used to detect the object

boundaries by using an edge detection operator and edge

information [28]. When there is abrupt change in intensity near

edge and little image noise then this method works better and it

uses convolving gradient operator [7]. First edges are identified

and linked close boundaries of the region to form required

boundaries. Some common edge detection operators are used

in this method such as Sobel operator, Laplace operator, Canny

operator, Laplace of Gaussian operator and so on. Canny edge

operator is most promising among all edge based operators but

it takes more time as compared to Sobel operator [1].The main

drawback of edge detection techniques is the presence of noise

that results in random variation in the level from pixel to pixel

[28].

Edge Detection algorithm are best suited for simple and noise

free images as well often produce missing edges or extra edges

on complex and noisy images [9 ]

2.3 Region based segmentation methods Based on the principle of homogeneity, neighboring pixel

having similar characteristic are grouped into one region and

dissimilar characteristics are in to other region. A region of an

image is defined as a connected homogeneous subset of the

image with respect to some criteria [1, 13]. This method is

simple and more immune to noise [4, 7]. Region based

methods are described below.

2.3.1 Region growing method Region growing is one of the most popular techniques for

segmentation due to simplicity and good performance. The

method groups pixels into larger regions based on predefined

criteria [10, 14]. It starts with a set of initial seed points and

grows the region. Seeds can be selected manually or

automatically. Automatic selection of seed points is based on

finding pixels that are of interest. Region growing algorithm is

presented below.

1. Select seed pixels within the image [4].

2. Select similarity criteria on the basis of grey level

intensity or color.

3. Appending each seed with neighboring pixels having

predefined properties similar to seed pixel in to the

region.

4. Repeat step iii) until no more pixels meet the criteria

for allocation into the regions.

The main drawback of region growing is that it requires

manual interaction to obtain the seed point. Region growing

can also be sensitive to noise, causing extracted regions to have

holes.

2.3.2 Region splitting and merging In contrast to region growing, this method works on the entire

image and segments the image on the basis of homogeneity

criteria [15]. The method utilizes quad tree data to distinguish

the homogeneity. It divides an image into the arbitrary

unconnected regions and then merges the regions on the basis

of condition of reasonable image segmentation [2, 7].

Initially, entire image is taken as a region and then the

technique divides the entire region into four quadrants on the

basis of predefined criteria. Repeating the process, it checks

each of the quadrants, and divides into four quadrants for the

same criteria. This procedure continues till the criteria satisfied

or no further division is possible. Figure 1 illustrates the

procedure.



3

Fig. 1. Quad tree

The algorithm is presented below.

Let R represents the entire image region and P a predicate

1. If [1], divides the image into

quadrants. If P is FALSE for any quadrants

i.e. then subdivide those

quadrants into sub quadrants and so on till no further

splitting is possible.

2. Merge any adjacent regions and for which,

.

3. Stop when no further merging is possible.

2.4 Segmentation methods based on PDE

(Partial Differential Equation) PDE image segmentation, introduced by Kassetal in 1987 [16],

is mainly carried out using active contour or snakes methods.

The method determines objects in presence of noise and

ambiguities. Snake method transforms a segmentation problem

into PDE framework and forwards to different methods of

image segmentation such as, snake, level set and Mumford-

Shah model.

2.4.1 Snake Method Active contours and snake segmentation methods are computer

generated curves which are used to find objects boundaries

under the internal and external forces [16, 17]. The work-in

procedure is described below.

1. Snake is placed near the contour of Region of

Interest (ROI)

2. Internal and External forces within the image control

the shape and location of the snake within image

[18].

3. The internal and external forces are used to calculate

the contour of ROI by constructing energy function.

The internal forces are used for smoothness while

external forces guide the contours towards the

contour of ROI.

The main drawback of snake method is that, it requires user

interaction for curve drawing around detected objects [18].

Apart from this drawback, snake algorithm is sensitive to noise

and computational complexity is high. Attempts have been

made to solve these problems [ 16 ].

2.4.2 Levelset Model This method developed by Osher and Sethian [16] is very

useful on moving curves and surfaces with curvature based

velocities. It represents the curves as the zero levelset of higher

dimensional hyper surface. The method provides accurate

numerical implementation and manages topological change.

Levelset procedure is described below.

1) Pick up a set of seed points that represent an initial

contour. It works for one seed point picked up

manually or automatically. But, the less the seed

points mask, the more the expensive calculations.

2) Create a signed distance function.

3) Compute feature image using gradient and gaussian

filter.

4) Get the curve’s narrow band.

5) Get curvature and use gradient descent to minimize

energy.

6) Evolve the curve.

7) Repeat the second step until obtaining the

segmented region.

The main advantage of levelset is its stability; solve the

problem of corner point producing, curve breaking and

combining etc. The main drawback of this method is that,

objects with edges defined by gradient can be segmented and

curve may pass through object boundaries.

2.5 Segmentation based on Artificial Neural

Network (ANN) ANNs are widely used in medical image segmentation in

recent years. Artificial representation of human brain that tries

to simulate its learning process is called as neural net [19, 20,

21]. The neural network is trained with the sample set and this

knowledge is used to segment the input image.

In this method, images are mapped into neural network. Image

segmentation problem is converted into energy minimization

problem where every neuron stands for a pixel [6, 7]. Neural

network segmentation consists of two steps based on neural

network- feature extraction and image segmentation. Some

features are extracted from the images, such that they are

suitable for segmentation and these are the inputs of the neural

network [17]. The basic structure of a neuron is described as

shown in Fig. 2 where represents the

input to the neurons and Y represents the output. Each input is

multiplied by its weight , a bias b is associated with each

neuron. Each neuron has one pixel in an input image, receiving

its corresponding pixels color information as an external

stimulus. Also each neuron connects with its neighboring

neurons, receiving local stimuli from them. The external and

local stimuli are combined in an internal activation system,

which accumulates the stimuli until it exceeds a dynamic

threshold, resulting in a pulse output. The temporary series of

pulse output shows information of input images and utilized for

many image processing application, such as image

segmentation and feature generation.

Y Output

Input Transfer function

Fig. 2. A typical ANN structure.

This segmentation method relies on processing small areas of

an image using artificial neural network or a set of neural

networks. Many different neural network architectures are

b

f



4

available such as Feedforward Network, Radial Basis Function

Network, Feedback Network, and Self Organizing Map [29].

The method reduces manual expert intervention during image

segmentation. The drawbacks of the method includes,

Some predefined information is needed before

segmentation.

Initialization process makes influence to the result of

image segmentation.

Neural network uses trained sample set using

learning process beforehand [7]. Therefore, period of

training may be long so we should avoid

overtraining.

2.6 Segmentation based on clustering Clustering is an unsupervised learning task. It mainly used to

identify a finite set of categories known as clusters to classify

pixels [22]. It is mainly used when advance knowledge of

classes is known. A predefined similarity criterion is defined

between pixels [2, 3]. Clusters are formed by grouping similar

pixels. The formation of cluster is based on principle of

maximizing the intra class similarity. Clustering is of different

types such as, hard clustering, k-means clustering, fuzzy

clustering etc.

2.6.1 Hard Clustering Hard clustering consists of accurate boundaries between

different clusters [23]. A pixel is belongs to only one cluster. A

well-known example of hard clustering algorithm is K-means

clustering algorithm [22, 23]. K-means clustering algorithm

partition n pixels in to k clusters, where k < n. K-means

algorithm is based on similarity and dissimilarity between pair

of data component. K- mean algorithm classify pixels in an

image into k number of clusters on the basis of similarity

feature like grey level intensity of pixels and distance of pixel

intensities from centroid pixel intensity[2]. Cluster can be

selected manually, randomly or based on some conditions.

Distance between the pixel and cluster center is measured by

the squared or absolute difference between a pixel and a cluster

center. This difference calculation is depends on pixel color,

intensity, texture and location or weighted combination of

these factors. The final output of the clustering method depends

on the initial set of clusters.

The main advantage of K-means algorithm is simple and

minimum execution cost. The main drawback of K-means

algorithm is the determination of K, number of cluster [24] and

it cannot give the same result every time when algorithm is

executed. Resulting cluster heavily depend on the initial

assignment of centroids.

2.6.2 Fuzzy Clustering It is very difficult to classify the pixel in an image correctly

[26] when there are no boundaries between objects in an

image. To remove this problem fuzzy clustering method is

used. Fuzzy clustering method classifies pixel values

accurately, so it is broadly used in decision oriented

applications such as tumor detection, tissue classification.

Fuzzy clustering classifies the pixels into clusters based on

similarity criteria i.e. distance; connectivity, intensity and

pixels belong to some cluster. Fuzzy clustering algorithm

includes FCM (Fuzzy C- Means), GK (Gustafson – kessel),

GMD (Gaussian mixture decomposition), FCV (Fuzzy C

varieties), AFC, FCS, FCSS, FCQS, FCRS etc.

3. EXPERIMENTAL RESULTS The number of techniques for medical image segmentation is

quite large and as a result focused on current techniques rather

than being exhaustive. The techniques described in previous

section are experimented on abdomen and liver CT images.

The results are presented below.

3.1 Thresholding

A threshold is used turn a grey-scale image into a binary

image. Threshold segmentation for the normal abdomen and

liver CT is shown in Fig. 3. A threshold value of 0.39 is chosen

on the basis of trial and error method. Threshold segmentation

of the CT abdomen and liver image consisting of abnormality

is shown in Fig. 4.

(a) Original image (b) Segmented image

Fig. 3. Threshold segmentation of normal CT image.


Fig. 4. Threshold segmentation of abnormal CT image.

Adaptive threshold segmentation of the normal and

abnormal abdomen and liver CT is shown in Fig. 5 and Fig. 6

respectively.


Fig. 5. Adaptive Threshold segmentation of normal CT

image.



5


Fig. 6. Adaptive Threshold segmentation of abnormal

CT image.

Otsu’s method: - Otsu’s method selects the threshold by

minimizing the within class variance of the two groups of

pixels separated by the thresholding operator which in turn

maximizes the between class variance. The result of applying

Otsu’s method for normal and abnormal abdomen and liver CT

are shown in Fig. 7 and Fig. 8.


Fig. 7. Otsu’s Threshold segmentation of normal CT

image.


Fig. 8. Otsu’s Threshold segmentation of abnormal CT

image.

3.2 Region Growing

This is a classical segmentation method. The seed mark each of

the objects to be segmented. Five seed point are used in

segmentation of the normal abdomen and liver CT images. The

result is shown in Fig. 9 and Fig. 10 for both normal and

abnormal CT images, respectively.


Fig. 9. Region growing segmentation of normal CT

image.


Fig. 10. Region growing segmentation of abnormal CT

image.

Region growing method gives better result as compared to the

thresholding method. Region growing clearly extract the

abnormality tissue area. However, the method requires manual

interaction for the selection of seed point.

3.3 Clustering The result of segmentation based on clustering is shown in Fig.

11 and Fig. 12 respectively.


Fig. 11. Clustering segmentation of normal CT image.

In clustering, there are many pixels missing as compared to the

threshold method and the region growing method. Missing of

pixels is due to noise interference and this lead to the holes in

the segmented image. The result in Fig. 12 clearly indicates the

abnormality part in abdomen and liver CT and satisfies the

clinical validation.



6


Fig. 12. Clustering segmentation of abnormal CT image.

3.4 Edge detection

Edge detection method detects the object boundaries by using

edge detection operator and extract boundaries by using the

edge information. Canny edge and Sobel operators are used for

edge detection. The results are shown in Fig. 13 and Fig. 14,

respectively for normal and abnormal CT images. The

complexity of medical images makes the correct boundary

detection very difficult.

(a)Original image (b) Canny edge (c) Sobel edge

Fig. 13. Edge Detection based segmentation of normal

CT image.

(a) Original image (b) Canny edge (c) Sobel

edge

Fig. 14. Edge Detection based segmentation of abnormal

CT image.

3.5 Level set based segmentation

Levelset method detects abnormality of abdomen and liver CT

by using levelset function. Levelset is a deformable contour

that is evolved to the image contour. Result of applying

Levelset method for normal and abnormal abdomen and liver

CT are shown in Fig. 15 and Fig. 16, respectively.


Fig. 15. Levelset based segmentation of normal CT

image.

(a) Original image (b) Segmented

image

Fig. 16. Levelset based segmentation of abnormal CT

image.

3.6 Region split and merging

Split and merge is a region based segmentation technique that

defines a predicate R which is the basis of splitting and

merging. The result of applying region split and merging

method for normal and abnormal abdomen and liver CT are

shown in Fig. 17 and Fig. 18, respectively.

(a) (b) (c )

Fig. 17. Region Split and Merge segmentation of normal

CT (a) Original image (b) normal segmented image (c)

segmented image.

(a) (b) ( c )

Fig. 18. Region Split and Merge segmentation of

abnormal CT (a) Original image (b) segmented image (c)

abnormal segmented image.



7

3.7 Chan-vese active contour

It performs active contour without edges on gray scale, binary

image with initial phase mask. In order to define a user mask,

the user should make sure that the dimensions of mask match

those of the input image. The result of applying Chan-vese

active contour method for normal and abnormal abdomen and

liver CT are shown in Fig. 19 and Fig. 20 respectively.

Fig. 19. Chen-vese segmentation of normal CT image.

Fig. 20. Chen-vese segmentation of abnormal CT image.

Mean, standard deviation, and pixel values are computed for

both normal and abnormal CT scan images. The values, correct

to 4 decimal places, are presented in TABLE I and TABLE II

for the segmentation methods studied.

TABLE I. parameter values for the segmented

normal abdomen and liver CT.

Sr.

No.

Method Mean Standard

Deviation

Pixel

Values

1 Thresholding 71.4425 79.1362 52836

2 Adaptive

thresholding 67.2618 46.6472 106072

3 Region growing 87.2836 32.8806 137647

4 Clustering 81.2850 38.3532 383795

5 Canny edge 8.7432 28.1380 13891

6 Sobel edge 3.1688 17.4709 5005

7 Levelset 43.1340 35.9951 67865

8 Region split and

merge 32.7954 144.1601 51816

9 Otsu’s method 76.3905 114.7954 421753

10 Chen-Vese active

contour 32.5973 46.5281 20912

TABLE II. Parameter values for the segmented

abnormal abdomen and liver CT (ROI).

Sr.

No.

Method Mean Standard

Deviation

Pixel

Values

1 Thresholding 74.1348 79.1988 55832

2 Adaptive

thresholding 68.8537 46.0686 109805

3 Region growing 86.5108 33.7665 138200

4 Clustering 81.6685 37.8246 390349

5 Canny edge 8.7915 28.2460 14183

6 Sobel edge 3.1071 17.3142 5000

7 Levelset 45.7904 36.8220 73224

8 Region split and

merge 31.8325 141.5739 51352

9 Otsu’s method 86.2060 118.5807 486085

10 Chen-Vese active

contour 35.1068 47.3675 22003

Further, mean computed using machine (Siemens somatom

scope 16 slice) were noted down for the same set of images

used for experimenting. The machine generated mean were

compared with mean obtained with our experimental results

and are tabulated in TABLE III and TABLE IV for normal and

affected images respectively. From the tables, it is evident that

Region growing method gave better results followed by Otsu’s

method compared to other segmentation methods.

TABLE III. Mean values measured for the normal

abdomen and liver CT by Machine.

Sr.

No.

Method Computed

Mean

Mean

measured

by

machine

Difference

Mean

1 Thresholding 71.4425 110 38.557

2 Adaptive

thresholding 67.2618 110 42.738

3 Region

growing 87.2836 110 22.716

4 Clustering 81.2850 110 28.715

5 Canny edge 8.7432 110 101.256

6 Sobel edge 3.1688 110 106.831

7 Levelset 43.1340 110 66.866

8 Region split

and merge 32.7954 110 77.204

9 Otsu’s method 76.3905 110 30.609

10 Chen-Vese

active contour 32.5973 110 77.402



8

TABLE IV. Mean values measured for the

abnormal abdomen and liver CT by Machine.

Sr.

No.

Method Computed

Mean

Mean

measured

by

machine

Difference

Mean

1 Thresholding 74.1348 114 39.865

2 Adaptive

thresholding 68.8537 114 45.146

3 Region growing 86.5108 114 27.489

4 Clustering 81.6685 114 32.332

5 Canny edge 8.7915 114 105.209

6 Sobel edge 3.1071 114 110.893

7 Levelset 45.7904 114 68.210

8 Region split and

merge 31.8325 114 82.168

9 Otsu’s method 86.2060 114 27.794

10 Chen-Vese active

contour 35.1068 114 78.893

4. CONCLUSION Image segmentation plays an important role in medical

imaging, by automating or providing the facility of the

delineation of anatomical structures and region of interest. In

this paper, several current segmentation techniques used in

medical imaging has been studied and experiments are

performed on liver and CT images for performance

comparison. Outcomes of the segmentation techniques which

are discussed in this paper depends on some input parameter

like threshold for the thresholding, number of cluster centers,

seed point for the region growing. Running time for the

clustering method depends on the number of iteration used and

edge detection depends on discontinuity of grey level or

intensity variation on the grey scale images. Region growing

needs manual interaction. If the threshold value is accurate then

thresholding gives best performance. From experimental

results it is evident that region growing method performed

better with results comparable with machine generated values

(mean value) than the other segmentation techniques. Our

future study is to combine existing techniques with

preprocessing procedures and volume estimation in order to

improve the results in segmentation.

5. REFERENCES [1] Rafael C. Gonzalez, Richard E. Woods, “Digital Image

Processing”, 2nd ed., Beijing: Publishing House of

Electronics Industry, 2007.

[2] S.Aksoy,“Image Segmentation”, Department of

Computer Engineering, Bilkent Univ.

[3] Zhang, Y. J, An Overview of Image and Video

Segmentation in the last 40 years, Proceedings of the 6th

International Symposium on Signal Processing and Its

Applications, pp. 144-151, 2001.

[4] K. K. Singh, A. Singh,“A Study of Image Segmentation

Algorithms for Different Types of Images”,

International Journal of Computer Science Issues, Vol. 7,

Issue 5, 2010.

[5] Jesmin F. Khan, Sharif M. A. Bhuiyan, and Reza R.

Adhami,” Image Segmentation and Shape Analysis for

Road-Sign Detection”, IEEE TRANSACTIONS ON

INTELLIGENT TRANSPORTATION SYSTEMS,

VOL. 12, NO. 1, MARCH 2011.

[6] N. R. Pal, S. K. Pal,“A Review on Image Segmentation

Techniques”, Pattern Recognition, Vol. 26, No. 9, pp.

1277- 1294, 1993.

[7] W. X. Kang, Q. Q. Yang, R. R. Liang,“The Comparative

Research on Image Segmentation Algorithms”, IEEE

Conference on ETCS, pp. 703-707, 2009.

[8] Rastgarpour M., and Shanbehzadeh J., Application of AI

Techniques in Medical Image Segmentation and Novel

Categorization of Available Methods and Tools,

Proceedings of the International MultiConference of

Engineers and Computer Scientists 2011 Vol I, IMECS

2011, March 16-18, 2011, Hong Kong.

[9] S. S. Varshney, N. Rajpal, R. Purwar,“Comparative

Study of Image Segmentation Techniques and Object

Matching using Segmentation”, Proceeding of

International Conference on Methods and Models in

Computer Science, pp. 1-6, 2009.

[10] K. G. Gunturk, “EE 7730- Image Analysis I”, Louisiana

state university.

[11] L.Aurdal,“Image Segmentation beyond thresholding”,

Norsk Regnescentral, 2006.

[12] Y. Zhang, H. Qu, Y. Wang,“Adaptive Image

Segmentation Based on Fast Thresholding and Image

Merging”, Artificial reality and Telexistence-Workshops,

pp. 308-311, 1994.

[13] H. G. Kaganami, Z. Beij,“Region Based Detection versus

Edge Detection”, IEEE Transactions on Intelligent

information hiding and multimedia signal processing, pp.

1217-1221, 2009.

[14] Y. Chang, X. Li,“Adaptive Image Region Growing”,

IEEE Trans. On Image Processing, Vol. 3, No. 6, 1994.

[15] Bo Peng, Lei Zhang, and David Zhang, “A Survey of

Graph Theoretical Approaches to Image Segmentation”.

[16] X. Jiang, R. Zhang, S. Nie,“Image Segmentation Based

on PDEs Model: a Survey”, IEEE conference, pp. 1-4,

2009.

[17] C. Zhu, J. Ni, Y. Li, G. Gu,“General Tendencies in

Segmentation of Medical Ultrasound Images”,

International Conference on ICICSE, pp. 113-117, 2009.

[18] P. Karch, I. Zolotova,“An Experimental Comparison of

Modern Methods of Segmentation”, IEEE 8th

International Symposium on SAMI, pp. 247-252, 2010.

[19] T.F. Wang, D.Y. Li et al. Automatic segmentation of

medical ultrasound image using self-creating and

organizing neural network.Journal of

electronics.1999,21(1),pp.124-127.

[20] Z. B. Chen, Q. H. Zheng, T. S. Qiu, Y. Liu. A new

method for medical ultrasonic image

segmentation.Chinese Journal of Biomedical

Engineering.2006, 25(6), pp.650-655.

[21] Y.L.Huang,D.R.Chen. Watershed segmentation for

breast tumor in 2Dsonography. Ultrasound in Medicine

& Biology. 2004, 30(5), pp.625632.



9

[22] V. K. Dehariya, S. K. Shrivastava, R. C. Jain,“Clustering

of Image Data Set Using K-Means and Fuzzy K-Means

Algorithms”, International conference on CICN, pp. 386-

391, 2010.

[23] F .Z. Kettaf, D. BI, J. P.,“A Comparison Study of Image

Segmentation by Clustering Technique”, Proceedings of

ICSP, pp. 1280-1282, 1996.

[24] P.Lukac, R. Hudec, M. Benco, P. Kamencay, Z.

Dubcova, M. Zachariasova,“Simple Comparison of

Image Segmentation Algorithms Based on Evaluation

Criterion”, IEEE Conference on Radioelektronika, pp. 1-

4, 2011.

[25] S.Tatiraju, A. Mehta,“Image Segmentation using k-

means clustering, EM and Normalized Cuts”,

Department of EECS, pp. 1-7.

[26] S. Naz, H. Majeed, H. Irshad,“Image Segmentation using

Fuzzy Clustering: A Survey”, International Conference

on ICET, pp.181-186, 2010.

[27] Dzung L. Pham, ChenyangXu, and Jerry L.

Princ,”Current Methods In Medical Image

Segmentation,” Department of Electrical and Computer

Engineering, The Johns Hopkins University,Annu. Rev.

Biomed. Eng. 2000. 02:315-37.

[28] Krit Somkantha, Nipon Theera-Umpon and Sansanee

Auephanwiriyakul,” Boundary Detectioon in Medical

Images Using Edge Following Algorithm Based on

Intencity Gradient and Texture Gradient Features” IEEE

transaction on biomedical engineering, vol. 58, no. 3,

march 2011.

[29] J.Jiang, p. Trundle and J. Ren, Digital Media and

Systems Research Institute, University of Bradford,”

Medical Image Analysis with Artificial Neural

Networks.”

[30] Rajeshwar Dass, Priyanka, Swapna devi,” Image

Segmentation Techniques”, IJECT vol. 3, march 2012.

[31] A.M Khan, Ravi S,” Image Segmentation Methods: A

Comparative Study” International Journal of Soft

Computing and Engineering (IJSCE), Vol-3, Issue-4,

September 2013.

[32] Christo Ananth, Karthika, Shivangi singh, Jennifer

Christa, Gracelyn Ida,” Graph Cutting Tumor Images”

International Journal of Advanced Research in

Computer Science and Software Engineering, vol 4, Issue

3, march 2014.

IJCATM : www.ijcaonline.org

Documents

Automatic Detection of Abnormalities Associated with ...€¦ · thresholding fails when background illumination is uneven. Multiple thresholds are used to avoid uneven illumination